* try to fix precision error on x86 platform step8.

Author: IRainman

include/fast_float/ascii_number.h

@@ -439,7 +439,7 @@ parse_number_string(UC const *p, UC const *pend,
     } else {
       // Now let's parse the explicit exponent.
       while ((p != pend) && is_integer(*p)) {
-        if (exp_number < 0x1000) {
+        if (exp_number < 0x10000000) {
           // check for exponent overflow if we have too many digits.
           UC const digit = UC(*p - UC('0'));
           exp_number = 10 * exp_number + static_cast<am_pow_t>(digit);

include/fast_float/bigint.h

@@ -39,7 +39,7 @@ constexpr limb_t bigint_limbs = bigint_bits / limb_bits;
 // vector-like type that is allocated on the stack. the entire
 // buffer is pre-allocated, and only the length changes.
 template <limb_t size> struct stackvec {
-  limb data[size] = {0};
+  limb data[size];
   // we never need more than 150 limbs
   uint_fast8_t length{0};
 

include/fast_float/float_common.h

@@ -446,7 +446,7 @@ typedef int_fast8_t am_bits_t;
 
 // Power bias is signed for handling a denormal float
 // or an invalid mantissa.
-typedef int32_t am_pow_t;
+typedef int_fast32_t am_pow_t;
 
 // Bias so we can get the real exponent with an invalid adjusted_mantissa.
 constexpr static am_pow_t invalid_am_bias = -0x8000;